0
Any-degree Qr factorization triangularizes polynomial systems
The unitary factor preserves the original solution set for zero- and positive-dimensional cases, extending numerical methods beyond linear a
full image
abstract click to expand
The unitary-triangular (QR) factorization of linear algebra may be used to robustly and efficiently solve a linear system. Toward a comparable numerical method to solve a polynomial system of higher degree, this paper proposes an any-degree unitary-triangular (Qr) factorization, which for a degree-one system reduces to the QR factorization. The work develops a tensor framework, i.e., codesigned algebra and software, where polynomial system coefficients are represented by a vector-shaped sparse tensor, a multidimensional array whose number of Ricci-notation indices, called the tensor degree, equals the highest monomial degree of the system. With the proposed Qr factorization, the coefficient tensor decomposes into a product of unitary and triangular factors that, in general, also have Ricci-notation indices and sparse entries. The unitary factor defines a unitary transform, a generalization of the related linear algebra concept to tensor algebra, that can triangularize a polynomial system while preserving its solution set, whether zero- or positive-dimensional. The work extends the author's Ricci-notation tensor framework, providing new algebra and new software to model, construct, and factorize polynomial systems in this manner. After applying the approach to numerically triangularize two zero-dimensional systems, chosen for educational value, results are compared to the Gr\"obner-basis (GB) method for triangularizing polynomial systems symbolically. One problem is of degree three, with three equations and unknowns, and the other of degree two, with four equations and unknowns. Although it resembles GB triangularization, the proposed Qr factorization has a completely different pedigree associated with numerical methods.
